Back to Work

AI & Automation

AI-Powered Video Generation Automation

Turns a topic into a ready-to-publish short video — through a live conversation, not a one-shot form.

A Telegram-driven pipeline that scripts, generates, narrates, and renders a short video from a single theme, with a review-and-revise checkpoint at every stage.

AI-Powered Video Generation Automation

Business Problem

What does it take to go from an idea to a finished short video?

Producing a short-form video normally means writing a script, recording or sourcing a voiceover, generating or shooting visuals, and editing everything together — a multi-tool process that's slow to repeat for every new idea, and hard to revise without redoing several steps by hand.

Business Impact

What Changes for Content Production

Conversational, Not a Form

Telegram-driven

A topic becomes a script, video, and voiceover through a live back-and-forth review, not a single unreviewable submission.

Revise Only What Changed

Each scene is individually addressable, so a revision can target one scene's video or voice instead of regenerating the whole video.

Fully Traceable

Every script, scene, and render is logged with its own status and history, so any run can be audited or resumed.

Suited For

Who this workflow fits.

Built for anyone who needs to turn a written idea into short-form video on a recurring basis, without operating a video editor by hand for every clip.

Solo Content Creators

Goes from a theme to a finished short video through a normal chat conversation.

Social Media & Marketing Teams

A repeatable script-to-video pipeline that supports revision requests instead of one-shot output.

Small Teams Publishing Regularly

Every stage is logged, so output can be tracked and reviewed before it goes live.

Automation-First Workflows

Each stage — script, video, voice, render, publish — is a swappable module, not a hard-coded chain to one provider.

If your team is manually stitching together AI scripts, generated visuals, and voiceovers, this pipeline automates that exact handoff.

Every stage — the script model, the video generator, the voice provider — can be swapped independently as better options appear.

Solution

From a Theme to a Finished Video

A Telegram conversation drives the entire pipeline: a submitted theme becomes an AI-written script, which the user can approve or revise in plain text. Once approved, each scene generates its own video and voiceover, gets rendered together, and comes back for a final review — with full or partial revisions supported — before publishing.

Theme
AI Script
Scene Video + Voice
Rendered Video
Publish

How It Works

Six steps, theme to publish.

  1. 01

    Submit

    User sends a video theme to the Telegram bot.

  2. 02

    Script

    AI generates a scene-by-scene script and sends it back for review.

  3. 03

    Review

    User approves or requests a free-text revision of the script.

  4. 04

    Generate

    Each scene's video and voiceover are generated.

  5. 05

    Render

    Scenes are combined into one video with synced narration.

  6. 06

    Publish

    User reviews the finished video and confirms it for publishing.

Architecture

Where automation ends and people decide.

Telegram Intake
Script Generation
Scene Video + Voice
Render
Human Review
Publish

Engineering Proof

What Makes the Workflow Robust?

Per-Scene Revision Addressing

Each scene has its own tracked state, so a revision can target one scene's video or voice without regenerating the entire video.

Modular, Swappable Providers

Script, video, voice, and render each run as an independent stage, so any single provider can be replaced without redesigning the pipeline.

Resilient to Empty Lookups

Stages that legitimately return no data are forced to still pass through, preventing the pipeline from silently stopping mid-run.

Idempotent Resume

Re-approving an already-approved step safely resumes the same run instead of starting a duplicate.

Validation

Found and Fixed Through Real Testing

  • A silent mid-run stall caused by an empty lookup result, traced and fixed
  • A stale node reference that broke the review notification, found via execution logs
  • A revision that updated the wrong data layer, corrected before it could generate from stale content
  • An unhandled status branch that misrouted follow-up messages, fixed and re-validated
  • Structural validation: 0 errors across every connection and expression in the workflow

My Role

AI Automation & Workflow Engineer

  • Workflow architecture
  • Conversational flow design
  • Multi-provider API integration
  • Per-scene state tracking
  • Error handling
  • Testing

Technology

Tools behind this system.

  • n8n
  • Gemini AI
  • Google Veo
  • ElevenLabs
  • Shotstack

Let's Work Together

Tell me about the process, and I'll tell you what can be automated or built.

[email protected]